The Classification Method for the Identification of Face Spoof in Convolutional Neural Networks

Abstract

Automatic facial recognition is currently extensively utilised in a variety of applications, ranging from identity deduplication to mobile payment verification. Face recognition has grown in popularity, raising worries about face spoof attacks (also known as biometric sensor presentation assaults), in which a picture or video of an authorised person's face may be used to obtain access to facilities or services without the person's knowledge. Even though a lot of face spoof detection methods have been suggested, their capacity to generalise has not been well investigated. On the basis of Image Distortion Analysis(IDA), we present an efficient and somewhat robust face spoof detection method . A new paradigm for each stage of a face recognition system is introduced in this article. In the phase of face identification, we present a hybrid model that combines AdaBoost and Convolutional Neural Network (ABCNN) to effectively handle the procedure. A multilayer perceptron and an active shape model will be used in conjunction with an ABANN to align the labelled faces identified in the previous phase. A mixture of Dense and Convolutional neural network layers was used to achieve binary classification of false recognition. The accuracy of categorial cross entropy prediction in Adam was found to be 91 percent, while the accuracy of SGD (stochastic gradient descent) was found to be 88 percent. In binary cross entropy, 90 percent accuracy was seen in Adam and 86 percent accuracy was observed in SGD, while in mean square, 86 percent accuracy was observed in Adam and 80 percent accuracy was observed in SGD.